Dataset Card for JL1-CD

Dataset Description (English)

Overview

JL1-CD is a large-scale, sub-meter, all-inclusive open-source dataset for remote sensing image change detection (CD). It contains 5,000 pairs of 512×512 pixel satellite images with a resolution of 0.5 to 0.75 meters, covering various types of surface changes in multiple regions of China. JL1-CD includes not only common human-induced changes (e.g., buildings, roads) but also natural changes (e.g., forests, water bodies, grasslands). The dataset aims to provide a comprehensive benchmark for change detection algorithms.

Dataset Structure

• Number of Image Pairs: 5,000 pairs
• Image Size: 512×512 pixels
• Resolution: 0.5 to 0.75 meters
• Change Types: Human-induced changes (e.g., buildings, roads) and natural changes (e.g., forests, water bodies, grasslands)
• Dataset Split: 4,000 pairs for training, 1,000 pairs for testing

Dataset Features

• High Resolution: Provides rich spatial information, facilitating visual interpretation.
• Comprehensive: Covers various change types, enhancing the generalization capability of algorithms.
• Open Source: The dataset is fully open-source, supporting the research community's use and improvement.

Usage

The JL1-CD dataset can be used to train and evaluate remote sensing image change detection models. The image pairs in the dataset include images from two time points along with corresponding pixel-level change labels. Users can utilize this dataset to develop, test, and optimize change detection algorithms.

Benchmark Results

The Multi-Teacher Knowledge Distillation (MTKD) framework proposed in the paper achieves new state-of-the-art (SOTA) results on the JL1-CD dataset. Experiments demonstrate that the MTKD framework significantly improves the performance of change detection models with various network architectures and parameter sizes.

Citation

If you use the JL1-CD dataset, please cite the following paper:

@article{liu2025jl1,
  title={JL1-CD: A New Benchmark for Remote Sensing Change Detection and a Robust Multi-Teacher Knowledge Distillation Framework},
  author={Liu, Ziyuan and Zhu, Ruifei and Gao, Long and Zhou, Yuanxiu and Ma, Jingyu and Gu, Yuantao},
  journal={arXiv preprint arXiv:2502.13407},
  year={2025}
}

License

The JL1-CD dataset is licensed under the MIT License, allowing users to freely use, modify, and distribute the dataset.

Contact

For any questions or suggestions, please contact:

• Ziyuan Liu: [email protected]

---

数据集描述 (中文)

概述

JL1-CD 是一个用于遥感图像变化检测（Change Detection, CD）的大规模、亚米级、全要素开源数据集。该数据集包含 5,000 对 512×512 像素的卫星图像，分辨率为 0.5 至 0.75 米，覆盖了中国多个地区的多种地表变化类型。JL1-CD 不仅包含常见的人为变化（如建筑物、道路），还涵盖了自然变化（如森林、水体、草地等）。该数据集旨在为变化检测算法提供一个全面的基准测试平台。

数据集结构

• 图像对数量：5,000 对
• 图像尺寸：512×512 像素
• 分辨率：0.5 至 0.75 米
• 变化类型：人为变化（如建筑物、道路）和自然变化（如森林、水体、草地）
• 数据集划分：4,000 对用于训练，1,000 对用于测试

数据集特点

• 高分辨率：提供丰富的空间信息，便于视觉解释。
• 综合性：涵盖多种变化类型，提升算法的泛化能力。
• 开源：数据集完全开源，支持研究社区的使用和改进。

使用方式

JL1-CD 数据集可用于训练和评估遥感图像变化检测模型。数据集中的图像对包含两个时间点的图像以及对应的像素级变化标签。用户可以使用该数据集来开发、测试和优化变化检测算法。

基准结果

论文中提出的 多教师知识蒸馏框架（MTKD） 在 JL1-CD 数据集上取得了最新的 state-of-the-art (SOTA) 结果。实验表明，MTKD 框架显著提升了多种网络架构和参数规模的变化检测模型的性能。

引用

如果您使用 JL1-CD 数据集，请引用以下论文：

@article{liu2025jl1,
  title={JL1-CD: A New Benchmark for Remote Sensing Change Detection and a Robust Multi-Teacher Knowledge Distillation Framework},
  author={Liu, Ziyuan and Zhu, Ruifei and Gao, Long and Zhou, Yuanxiu and Ma, Jingyu and Gu, Yuantao},
  journal={arXiv preprint arXiv:2502.13407},
  year={2025}
}

许可证

JL1-CD 数据集采用 MIT License，允许用户自由使用、修改和分发数据集。

联系方式

如有任何问题或建议，请联系：

• 刘子源: [email protected]